Intelligent evaluation method for identifying favorable shale oil areas based on improved stacked sparse autoencoder

Abstract

Identifying and evaluating favorable areas is crucial for shale oil exploration and development, well-location deployment, and fracturing design. Traditional machine learning methods struggle to accurately extract the characteristics of favorable shale oil areas with limited labeled data, affecting accuracy and generalization. This study proposes an intelligent method for identifying favorable shale oil areas under semi-supervised learning (SSAE-plus) to identify and evaluate favorable shale oil areas of the Qingshankou Formation in the Songliao Basin. The experimental results show that this method can effectively overcome the favorable area identification model’s reliance on labeled data and can adaptively extract the characteristics of favorable shale oil areas without supervision. The accuracy of model identification is as high as 98.82 percent. Compared with other methods, the SSAE-plus yields higher accuracy and efficiency, while being more stable and generalizable. The SSAE-plus achieved over 95 percent accuracy in identifying favorable shale oil areas across six datasets. It has broad application prospects in identifying and evaluating favorable areas, and provides valuable theoretical insights for shale oil development and exploration well layout.

Keywords

• shale oil
• favorable area
• autoencoder
• batch normalization
• semi-supervised learning